metag: Add NPTL support

Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

44 files changed, 1975 insertions, 95 deletions

diff --git a/include/elf.h b/include/elf.h
index 877d08194..cbb4da218 100644
--- a/include/elf.h
+++ b/include/elf.h
@@ -3184,6 +3184,11 @@ typedef Elf32_Addr Elf32_Conflict;
 #define R_METAG_RELATIVE                45
 #define R_METAG_GLOB_DAT                46
 
+/* TLS relocations */
+#define R_METAG_TLS_TPOFF               56
+#define R_METAG_TLS_DTPMOD              57
+#define R_METAG_TLS_DTPOFF              58
+
 #ifdef	__cplusplus
 }
 #endif
diff --git a/ldso/ldso/metag/dl-debug.h b/ldso/ldso/metag/dl-debug.h
index 5981d7c73..46c257c5c 100644
--- a/ldso/ldso/metag/dl-debug.h
+++ b/ldso/ldso/metag/dl-debug.h
@@ -17,11 +17,17 @@ static const char *_dl_reltypes_tab[] = {
 	[6]  "R_METAG_REG32OP1", "R_METAG_REG32OP2", "R_METAG_REG32OP3",
 	[9]  "R_METAG_REG16OP1", "R_METAG_REG16OP2", "R_METAG_REG16OP3",
 	[12] "R_METAG_REG32OP4", "R_METAG_HIOG", "R_METAG_LOOG",
-	[30] "R_METAG_VTINHERIT", "R_METAG_VTENTRY",
+	[30] "R_METAG_GNU_VTINHERIT", "R_METAG_GNU_VTENTRY",
 	[32] "R_METAG_HI16_GOTOFF", "R_METAG_LO16_GOTOFF",
 	[34] "R_METAG_GETSET_GOTOFF", "R_METAG_GETSET_GOT",
 	[36] "R_METAG_HI16_GOTPC", "R_METAG_LO16_GOTPC",
 	[38] "R_METAG_HI16_PLT", "R_METAG_LO16_PLT",
 	[40] "R_METAG_RELBRANCH_PLT", "R_METAG_GOTOFF",
 	[42] "R_METAG_PLT", "R_METAG_COPY", "R_METAG_JMP_SLOT",
+	[45] "R_METAG_RELATIVE", "R_METAG_GLOB_DAT", "R_METAG_TLS_GD",
+	[48] "R_METAG_TLS_LDM", "R_METAG_TLS_LDO_HI16", "R_METAG_TLS_LDO_LO16",
+	[51] "R_METAG_TLS_LDO", "R_METAG_TLS_IE", "R_METAG_TLS_IENONPIC",
+	[54] "R_METAG_TLS_IENONPIC_HI16", "R_METAG_TLS_IENONPIC_LO16",
+	[56] "R_METAG_TLS_TPOFF", "R_METAG_TLS_DTPMOD", "R_METAG_TLS_DTPOFF",
+	[59] "R_METAG_TLS_LE", "R_METAG_TLS_LE_HI16", "R_METAG_TLS_LE_LO16"
 };
diff --git a/ldso/ldso/metag/dl-startup.h b/ldso/ldso/metag/dl-startup.h
index 8dbf747e1..32b2e4b74 100644
--- a/ldso/ldso/metag/dl-startup.h
+++ b/ldso/ldso/metag/dl-startup.h
@@ -21,6 +21,11 @@ __asm__ (
 "       CALLR   D1RtP,__dl_start\n"
 "       GETL    D0Ar2,D1Ar1,[A0StP+#-(1*8)]\n"
 "       GETL    D0Ar4,D1Ar3,[A0StP+#-(2*8)]\n"
+"       ADDT    A1LbP,CPC1,#HI(__GLOBAL_OFFSET_TABLE__)\n"
+"       ADD     A1LbP,A1LbP,#LO(__GLOBAL_OFFSET_TABLE__+4)\n"
+"       ADDT    A1LbP,A1LbP,#HI(__dl_fini@GOTOFF)\n"
+"       ADD     A1LbP,A1LbP,#LO(__dl_fini@GOTOFF)\n"
+"       MOV     D0Ar4, A1LbP\n"
 "       SUB     A0StP,A0StP,#(2*8)\n"
 "       MOV     PC,D0Re0\n"
 "       .size __start,.-__start\n"
diff --git a/ldso/ldso/metag/dl-sysdep.h b/ldso/ldso/metag/dl-sysdep.h
index 29547cb34..257ca610f 100644
--- a/ldso/ldso/metag/dl-sysdep.h
+++ b/ldso/ldso/metag/dl-sysdep.h
@@ -36,14 +36,17 @@ extern unsigned long _dl_linux_resolver(struct elf_resolve *tpnt, int reloc_entr
 #define PAGE_ALIGN (~ADDR_ALIGN)
 #define OFFS_ALIGN (PAGE_ALIGN & ~(1ul << (sizeof(_dl_pagesize) * 8 - 1)))
 
-/* The union of reloc-type-classes where the reloc TYPE is a member.
-
-   TYPE is in the class ELF_RTYPE_CLASS_NOCOPY if it should not be allowed
-   to resolve to one of the main executable's symbols, as for a COPY
-   reloc.  */
-#define elf_machine_type_class(type)				\
-  (((((type) == R_METAG_JMP_SLOT)) * ELF_RTYPE_CLASS_PLT)	\
-   | (((type) == R_METAG_COPY) * ELF_RTYPE_CLASS_COPY))
+/* ELF_RTYPE_CLASS_PLT iff TYPE describes relocation of a PLT entry or
+   TLS variable, so undefined references should not be allowed to
+   define the value.
+
+   ELF_RTYPE_CLASS_NOCOPY iff TYPE should not be allowed to resolve to one
+   of the main executable's symbols, as for a COPY reloc.  */
+#define elf_machine_type_class(type)                                 \
+  ((((type) == R_METAG_JMP_SLOT || (type) == R_METAG_TLS_DTPMOD      \
+     || (type) == R_METAG_TLS_DTPOFF || (type) == R_METAG_TLS_TPOFF) \
+     * ELF_RTYPE_CLASS_PLT)                                          \
+     | (((type) == R_METAG_COPY) * ELF_RTYPE_CLASS_COPY))
 
 static inline Elf32_Addr
 elf_machine_dynamic(Elf32_Ehdr *header)
diff --git a/ldso/ldso/metag/elfinterp.c b/ldso/ldso/metag/elfinterp.c
index 78434167b..e0f981741 100644
--- a/ldso/ldso/metag/elfinterp.c
+++ b/ldso/ldso/metag/elfinterp.c
@@ -43,43 +43,36 @@ static inline void __put_unaligned_reloc(unsigned long *addr,
 unsigned long
 _dl_linux_resolver(struct elf_resolve *tpnt, int reloc_entry)
 {
-	int reloc_type;
 	int symtab_index;
 	char *strtab;
 	char *symname;
 	char *new_addr;
 	char *rel_addr;
 	char **got_addr;
-	Elf32_Sym *symtab;
+	ElfW(Sym) *symtab;
 	ELF_RELOC *this_reloc;
 	unsigned long instr_addr;
 
 	rel_addr = (char *)tpnt->dynamic_info[DT_JMPREL];
 
 	this_reloc = (ELF_RELOC *)(intptr_t)(rel_addr + reloc_entry);
-	reloc_type = ELF32_R_TYPE(this_reloc->r_info);
-	symtab_index = ELF32_R_SYM(this_reloc->r_info);
+	symtab_index = ELF_R_SYM(this_reloc->r_info);
 
-	symtab = (Elf32_Sym *)(intptr_t)tpnt->dynamic_info[DT_SYMTAB];
+	symtab = (ElfW(Sym) *)(intptr_t)tpnt->dynamic_info[DT_SYMTAB];
 	strtab = (char *)tpnt->dynamic_info[DT_STRTAB];
 	symname = strtab + symtab[symtab_index].st_name;
 
-	if (unlikely(reloc_type != R_METAG_JMP_SLOT)) {
-		_dl_dprintf(2, "%s: Incorrect relocation type in jump relocations\n",
-			    _dl_progname);
-		_dl_exit(1);
-	}
-
 	/* Address of the jump instruction to fix up. */
 	instr_addr = ((unsigned long)this_reloc->r_offset +
 		      (unsigned long)tpnt->loadaddr);
 	got_addr = (char **)instr_addr;
 
 	/* Get the address of the GOT entry. */
-	new_addr = _dl_find_hash(symname, tpnt->symbol_scope, tpnt,
+	new_addr = _dl_find_hash(symname, &_dl_loaded_modules->symbol_scope, tpnt,
 			ELF_RTYPE_CLASS_PLT, NULL);
 	if (unlikely(!new_addr)) {
-		_dl_dprintf(2, "%s: Can't resolve symbol '%s'\n", _dl_progname, symname);
+		_dl_dprintf(2, "%s: Can't resolve symbol '%s'\n",
+			    _dl_progname, symname);
 		_dl_exit(1);
 	}
 
@@ -102,28 +95,28 @@ _dl_linux_resolver(struct elf_resolve *tpnt, int reloc_entry)
 }
 
 static int
-_dl_parse(struct elf_resolve *tpnt, struct dyn_elf *scope,
+_dl_parse(struct elf_resolve *tpnt, struct r_scope_elem *scope,
 	  unsigned long rel_addr, unsigned long rel_size,
-	  int (*reloc_fnc)(struct elf_resolve *tpnt, struct dyn_elf *scope,
-			   ELF_RELOC *rpnt, Elf32_Sym *symtab, char *strtab))
+	  int (*reloc_fnc)(struct elf_resolve *tpnt, struct r_scope_elem *scope,
+			   ELF_RELOC *rpnt, ElfW(Sym) *symtab, char *strtab))
 {
 	int symtab_index;
 	unsigned int i;
 	char *strtab;
-	Elf32_Sym *symtab;
+	ElfW(Sym) *symtab;
 	ELF_RELOC *rpnt;
 
 	/* Parse the relocation information. */
 	rpnt = (ELF_RELOC *)(intptr_t)rel_addr;
 	rel_size /= sizeof(ELF_RELOC);
 
-	symtab = (Elf32_Sym *)(intptr_t)tpnt->dynamic_info[DT_SYMTAB];
+	symtab = (ElfW(Sym) *)(intptr_t)tpnt->dynamic_info[DT_SYMTAB];
 	strtab = (char *)tpnt->dynamic_info[DT_STRTAB];
 
 	for (i = 0; i < rel_size; i++, rpnt++) {
 		int res;
 
-		symtab_index = ELF32_R_SYM(rpnt->r_info);
+		symtab_index = ELF_R_SYM(rpnt->r_info);
 
 		debug_sym(symtab, strtab, symtab_index);
 		debug_reloc(symtab, strtab, rpnt);
@@ -141,7 +134,7 @@ _dl_parse(struct elf_resolve *tpnt, struct dyn_elf *scope,
 				    strtab + symtab[symtab_index].st_name);
 
 		if (unlikely(res < 0)) {
-			int reloc_type = ELF32_R_TYPE(rpnt->r_info);
+			int reloc_type = ELF_R_TYPE(rpnt->r_info);
 
 #if defined (__SUPPORT_LD_DEBUG__)
 			_dl_dprintf(2, "can't handle reloc type %s\n",
@@ -161,8 +154,8 @@ _dl_parse(struct elf_resolve *tpnt, struct dyn_elf *scope,
 }
 
 static int
-_dl_do_reloc(struct elf_resolve *tpnt, struct dyn_elf *scope,
-	     ELF_RELOC *rpnt, Elf32_Sym *symtab, char *strtab)
+_dl_do_reloc(struct elf_resolve *tpnt, struct r_scope_elem *scope,
+	     ELF_RELOC *rpnt, ElfW(Sym) *symtab, char *strtab)
 {
 	int reloc_type;
 	int symtab_index;
@@ -170,30 +163,35 @@ _dl_do_reloc(struct elf_resolve *tpnt, struct dyn_elf *scope,
 	unsigned long *reloc_addr;
 	unsigned long symbol_addr;
 #if defined (__SUPPORT_LD_DEBUG__)
-	unsigned long old_val;
+	unsigned long old_val = 0;
 #endif
+	struct elf_resolve *tls_tpnt = NULL;
+	struct symbol_ref sym_ref;
 
-	reloc_addr = (unsigned long *)(intptr_t)(tpnt->loadaddr + (unsigned long)rpnt->r_offset);
-	reloc_type = ELF32_R_TYPE(rpnt->r_info);
-	symtab_index = ELF32_R_SYM(rpnt->r_info);
+	reloc_addr = (unsigned long *)(tpnt->loadaddr + rpnt->r_offset);
+	reloc_type = ELF_R_TYPE(rpnt->r_info);
+	symtab_index = ELF_R_SYM(rpnt->r_info);
 	symbol_addr = 0;
-	symname = strtab + symtab[symtab_index].st_name;
+	sym_ref.sym = &symtab[symtab_index];
+	sym_ref.tpnt = NULL;
 
 	if (symtab_index) {
-		if (symtab[symtab_index].st_shndx != SHN_UNDEF &&
-		    ELF32_ST_BIND(symtab[symtab_index].st_info) == STB_LOCAL) {
-			symbol_addr = (unsigned long)tpnt->loadaddr;
-		} else {
-		  symbol_addr = (unsigned long)_dl_find_hash(symname, scope, tpnt,
-							     elf_machine_type_class(reloc_type), NULL);
-		}
-
-		if (unlikely(!symbol_addr && ELF32_ST_BIND(symtab[symtab_index].st_info) != STB_WEAK)) {
-			_dl_dprintf(2, "%s: can't resolve symbol '%s'\n", _dl_progname, symname);
-			_dl_exit(1);
+		symname = strtab + symtab[symtab_index].st_name;
+		symbol_addr = (unsigned long)_dl_find_hash(symname, scope, tpnt,
+							   elf_machine_type_class(reloc_type), &sym_ref);
+
+		if (!symbol_addr
+		    && ELF_ST_TYPE(symtab[symtab_index].st_info) != STT_TLS
+		    && ELF_ST_BIND(symtab[symtab_index].st_info) != STB_WEAK) {
+			_dl_dprintf(2, "%s: can't resolve symbol '%s'\n",
+				    _dl_progname, symname);
+			return 1;
 		};
-
-		symbol_addr += rpnt->r_addend;
+		if (_dl_trace_prelink) {
+			_dl_debug_lookup(symname, tpnt, &symtab[symtab_index],
+				&sym_ref, elf_machine_type_class(reloc_type));
+		}
+		tls_tpnt = sym_ref.tpnt;
 	}
 
 #if defined (__SUPPORT_LD_DEBUG__)
@@ -201,13 +199,21 @@ _dl_do_reloc(struct elf_resolve *tpnt, struct dyn_elf *scope,
 		old_val = __get_unaligned_reloc(reloc_addr);
 #endif
 
+#if defined USE_TLS && USE_TLS
+	/* In case of a TLS reloc, tls_tpnt NULL means we have an 'anonymous'
+	   symbol.  This is the case for a static tls variable, so the lookup
+	   module is just that one is referencing the tls variable. */
+	if (!tls_tpnt)
+		tls_tpnt = tpnt;
+#endif
 	switch (reloc_type) {
 		case R_METAG_NONE:
 			break;
 		case R_METAG_GLOB_DAT:
 		case R_METAG_JMP_SLOT:
 		case R_METAG_ADDR32:
-			__put_unaligned_reloc(reloc_addr, symbol_addr);
+			__put_unaligned_reloc(reloc_addr,
+					      symbol_addr + rpnt->r_addend);
 			break;
 		case R_METAG_COPY:
 #if defined (__SUPPORT_LD_DEBUG__)
@@ -215,11 +221,12 @@ _dl_do_reloc(struct elf_resolve *tpnt, struct dyn_elf *scope,
 				_dl_dprintf(_dl_debug_file,
 					    "\t%s move %d bytes from %x to %x\n",
 					    symname, symtab[symtab_index].st_size,
-					    symbol_addr, reloc_addr);
+					    symbol_addr + rpnt->r_addend,
+					    reloc_addr);
 #endif
 
 			_dl_memcpy((char *)reloc_addr,
-				   (char *)symbol_addr,
+				   (char *)symbol_addr + rpnt->r_addend,
 				   symtab[symtab_index].st_size);
 			break;
 		case R_METAG_RELATIVE:
@@ -227,13 +234,24 @@ _dl_do_reloc(struct elf_resolve *tpnt, struct dyn_elf *scope,
 					      (unsigned long)tpnt->loadaddr +
 					      rpnt->r_addend);
 			break;
+#if defined USE_TLS && USE_TLS
+		case R_METAG_TLS_DTPMOD:
+			*reloc_addr = tls_tpnt->l_tls_modid;
+			break;
+		case R_METAG_TLS_DTPOFF:
+			*reloc_addr = symbol_addr;
+			break;
+		case R_METAG_TLS_TPOFF:
+			CHECK_STATIC_TLS ((struct link_map *) tls_tpnt);
+			*reloc_addr = tls_tpnt->l_tls_offset + symbol_addr + rpnt->r_addend;
+			break;
+#endif
 		default:
 			return -1;	/* Calls _dl_exit(1). */
 	}
 
 #if defined (__SUPPORT_LD_DEBUG__)
-	if (_dl_debug_reloc && _dl_debug_detail &&
-	    (reloc_type != R_METAG_NONE)) {
+	if (_dl_debug_reloc && _dl_debug_detail && reloc_type != R_METAG_NONE) {
 		unsigned long new_val = __get_unaligned_reloc(reloc_addr);
 		_dl_dprintf(_dl_debug_file, "\tpatched: %x ==> %x @ %x\n",
 			    old_val, new_val, reloc_addr);
@@ -244,8 +262,8 @@ _dl_do_reloc(struct elf_resolve *tpnt, struct dyn_elf *scope,
 }
 
 static int
-_dl_do_lazy_reloc(struct elf_resolve *tpnt, struct dyn_elf *scope,
-		  ELF_RELOC *rpnt, Elf32_Sym *symtab, char *strtab)
+_dl_do_lazy_reloc(struct elf_resolve *tpnt, struct r_scope_elem *scope,
+		  ELF_RELOC *rpnt, ElfW(Sym) *symtab, char *strtab)
 {
 	int reloc_type;
 	unsigned long *reloc_addr;
@@ -253,8 +271,8 @@ _dl_do_lazy_reloc(struct elf_resolve *tpnt, struct dyn_elf *scope,
 	unsigned long old_val;
 #endif
 
-	reloc_addr = (unsigned long *)(intptr_t)(tpnt->loadaddr + (unsigned long)rpnt->r_offset);
-	reloc_type = ELF32_R_TYPE(rpnt->r_info);
+	reloc_addr = (unsigned long *)(tpnt->loadaddr + rpnt->r_offset);
+	reloc_type = ELF_R_TYPE(rpnt->r_info);
 
 #if defined (__SUPPORT_LD_DEBUG__)
 	old_val = *reloc_addr;
@@ -291,9 +309,10 @@ _dl_parse_lazy_relocation_information(struct dyn_elf *rpnt,
 
 int
 _dl_parse_relocation_information(struct dyn_elf *rpnt,
+				 struct r_scope_elem *scope,
 				 unsigned long rel_addr,
 				 unsigned long rel_size)
 {
-	return _dl_parse(rpnt->dyn, rpnt->dyn->symbol_scope, rel_addr,
+	return _dl_parse(rpnt->dyn, scope, rel_addr,
 			 rel_size, _dl_do_reloc);
 }
diff --git a/ldso/ldso/metag/metag_load_tp.S b/ldso/ldso/metag/metag_load_tp.S
new file mode 100644
index 000000000..2f00a9fef
--- /dev/null
+++ b/ldso/ldso/metag/metag_load_tp.S
@@ -0,0 +1,20 @@
+!    Copyright (C) 2013 Imagination Technologies Ltd.
+
+!    Licensed under LGPL v2.1 or later, see the file COPYING.LIB in this tarball.
+
+#include <features.h>
+
+#ifdef __UCLIBC_HAS_THREADS_NATIVE__
+
+#include <sysdep.h>
+
+	.text
+	.global ___metag_load_tp
+	.type   ___metag_load_tp,@function
+
+___metag_load_tp:
+	MOVT	D1Ar1,#HI(0x6ffff000)
+	JUMP	D1Ar1,#LO(0x6ffff000)
+	.size   ___metag_load_tp,.-___metag_load_tp
+
+#endif /* __UCLIBC_HAS_THREADS_NATIVE__ */
diff --git a/libc/sysdeps/linux/common/Makefile.in b/libc/sysdeps/linux/common/Makefile.in
index f7083a48b..dbf0b0fd4 100644
--- a/libc/sysdeps/linux/common/Makefile.in
+++ b/libc/sysdeps/linux/common/Makefile.in
@@ -69,6 +69,7 @@ CSRC- += fork.c getpid.c raise.c #open.c close.c read.c write.c
 CSRC- += $(if $(findstring =arm=,=$(TARGET_ARCH)=),vfork.c)
 CSRC- += $(if $(findstring =x86_64=,=$(TARGET_ARCH)=),vfork.c)
 #CSRC- += $(if $(findstring =mips=y=,=$(TARGET_ARCH)=$(CONFIG_MIPS_O32_ABI)=),waitpid.c)
+CSRC- += $(if $(findstring =metag=,=$(TARGET_ARCH)=),vfork.c)
 endif
 ifneq ($(ARCH_HAS_DEPRECATED_SYSCALLS),y)
 # No conversion is needed for new architectures
diff --git a/libc/sysdeps/linux/metag/Makefile.arch b/libc/sysdeps/linux/metag/Makefile.arch
index 0e6fbfe12..3f11d6da8 100644
--- a/libc/sysdeps/linux/metag/Makefile.arch
+++ b/libc/sysdeps/linux/metag/Makefile.arch
@@ -7,4 +7,6 @@
 
 CSRC-y := brk.c syscall.c metag.c __syscall_error.c
 
-SSRC-y := _longjmp.S clone.S setjmp.S
+SSRC-y := _longjmp.S clone.S setjmp.S vfork.S
+
+SSRC-$(UCLIBC_HAS_THREADS_NATIVE) += libc-metag_load_tp.S
diff --git a/libc/sysdeps/linux/metag/bits/atomic.h b/libc/sysdeps/linux/metag/bits/atomic.h
new file mode 100644
index 000000000..64aa50bc4
--- /dev/null
+++ b/libc/sysdeps/linux/metag/bits/atomic.h
@@ -0,0 +1,66 @@
+/*
+ * Copyrith (C) 2013 Imagination Technologies Ltd.
+ *
+ * Licensed under LGPL v2.1 or later, see the file COPYING.LIB in this tarball.
+ *
+ */
+
+#include <stdint.h>
+#include <sysdep.h>
+
+typedef int8_t atomic8_t;
+typedef uint8_t uatomic8_t;
+typedef int_fast8_t atomic_fast8_t;
+typedef uint_fast8_t uatomic_fast8_t;
+
+typedef int32_t atomic32_t;
+typedef uint32_t uatomic32_t;
+typedef int_fast32_t atomic_fast32_t;
+typedef uint_fast32_t uatomic_fast32_t;
+
+typedef intptr_t atomicptr_t;
+typedef uintptr_t uatomicptr_t;
+typedef intmax_t atomic_max_t;
+typedef uintmax_t uatomic_max_t;
+
+void __metag_link_error (void);
+
+#define atomic_full_barrier() \
+     __asm__ __volatile__("": : :"memory")
+
+/* Atomic compare and exchange.  This sequence relies on the kernel to
+   provide a compare and exchange operation which is atomic. */
+
+#define __arch_compare_and_exchange_val_8_acq(mem, newval, oldval) \
+  ({ __metag_link_error (); oldval; })
+
+#define __arch_compare_and_exchange_val_16_acq(mem, newval, oldval) \
+  ({ __metag_link_error (); oldval; })
+
+/* This code uses the kernel helper to do cmpxchg. It relies on the fact
+   the helper code only clobbers D0Re0. */
+#define __arch_compare_and_exchange_val_32_acq(mem, newval, oldval)     \
+  ({ register __typeof (oldval) a_current __asm__ ("D1Ar1");		\
+    register __typeof (oldval) a_newval __asm__ ("D0Ar2") = (newval);   \
+    register __typeof (mem) a_ptr __asm__ ("D1Ar3") = (mem);            \
+    register __typeof (oldval) a_oldval __asm__ ("D0Ar4") = (oldval);	\
+    __asm__ __volatile__						\
+      ("0:\n\t"								\
+       "GETD	%[cur], [%[ptr]]\n\t"					\
+       "CMP	%[cur], %[old]\n\t"					\
+       "BNE	1f\n\t"							\
+       "MOVT	D1RtP, #0x6fff\n\t"					\
+       "ADD	D1RtP, D1RtP, #0xf040\n\t"				\
+       "SWAP	D1RtP, PC\n\t"						\
+       "MOV	%[cur], %[old]\n\t"					\
+       "CMP	D0Re0, #0\n\t"						\
+       "BNE	0b\n\t"							\
+       "1:"								\
+       : [cur] "=&r" (a_current)					\
+       : [new] "r" (a_newval), [ptr] "r" (a_ptr),			\
+	 [old] "r" (a_oldval)						\
+       : "D0Re0", "D1RtP", "cc", "memory");				\
+    a_current; })
+
+#define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
+  ({ __metag_link_error (); oldval; })
diff --git a/libc/sysdeps/linux/metag/bits/syscalls.h b/libc/sysdeps/linux/metag/bits/syscalls.h
index b5c8fc58c..7ea09c2c3 100644
--- a/libc/sysdeps/linux/metag/bits/syscalls.h
+++ b/libc/sysdeps/linux/metag/bits/syscalls.h
@@ -49,8 +49,9 @@
 (__extension__                                                          \
   ({unsigned int __sys_result;						\
     {									\
+      PREP_ARGS_##nr (args);						\
       register int _result __asm__ ("D0Re0"), _nr __asm__ ("D1Re0");	\
-      LOAD_ARGS_##nr (args);						\
+      LOAD_ARGS_##nr;							\
       _nr = (name);							\
       __asm__ volatile ("SWITCH  #0x440001 ! syscall " #name		\
 			: "=r" (_result)				\
@@ -68,32 +69,52 @@
 #undef INTERNAL_SYSCALL_ERRNO
 #define INTERNAL_SYSCALL_ERRNO(val, err)	(-(val))
 
-#define LOAD_ARGS_0()
+#define PREP_ARGS_0()
+#define PREP_ARGS_1(a1)                                 \
+  int _t1 = (int) (a1);                                 \
+  PREP_ARGS_0 ()
+#define PREP_ARGS_2(a1, a2)                             \
+  int _t2 = (int) (a2);                                 \
+  PREP_ARGS_1 (a1)
+#define PREP_ARGS_3(a1, a2, a3)                         \
+  int _t3 = (int) (a3);                                 \
+  PREP_ARGS_2 (a1, a2)
+#define PREP_ARGS_4(a1, a2, a3, a4)                     \
+  int _t4 = (int) (a4);                                 \
+  PREP_ARGS_3 (a1, a2, a3)
+#define PREP_ARGS_5(a1, a2, a3, a4, a5)                 \
+  int _t5 = (int) (a5);                                 \
+  PREP_ARGS_4 (a1, a2, a3, a4)
+#define PREP_ARGS_6(a1, a2, a3, a4, a5, a6)             \
+  int _t6 = (int) (a6);                                 \
+  PREP_ARGS_5 (a1, a2, a3, a4, a5)
+
+#define LOAD_ARGS_0
 #define ASM_ARGS_0
-#define LOAD_ARGS_1(a1)					\
-  register int _a1 __asm__ ("D1Ar1") = (int) (a1);	\
-  LOAD_ARGS_0 ()
-#define ASM_ARGS_1	ASM_ARGS_0, "d" (_a1)
-#define LOAD_ARGS_2(a1, a2)				\
-  register int _a2 __asm__ ("D0Ar2") = (int) (a2);	\
-  LOAD_ARGS_1 (a1)
-#define ASM_ARGS_2	ASM_ARGS_1, "d" (_a2)
-#define LOAD_ARGS_3(a1, a2, a3)				\
-  register int _a3 __asm__ ("D1Ar3") = (int) (a3);	\
-  LOAD_ARGS_2 (a1, a2)
-#define ASM_ARGS_3	ASM_ARGS_2, "d" (_a3)
-#define LOAD_ARGS_4(a1, a2, a3, a4)			\
-  register int _a4 __asm__ ("D0Ar4") = (int) (a4);	\
-  LOAD_ARGS_3 (a1, a2, a3)
-#define ASM_ARGS_4	ASM_ARGS_3, "d" (_a4)
-#define LOAD_ARGS_5(a1, a2, a3, a4, a5)			\
-  register int _a5 __asm__ ("D1Ar5") = (int) (a5);	\
-  LOAD_ARGS_4 (a1, a2, a3, a4)
-#define ASM_ARGS_5	ASM_ARGS_4, "d" (_a5)
-#define LOAD_ARGS_6(a1, a2, a3, a4, a5, a6)		\
-  register int _a6 __asm__ ("D0Ar6") = (int) (a6);	\
-  LOAD_ARGS_5 (a1, a2, a3, a4, a5)
-#define ASM_ARGS_6	ASM_ARGS_5, "d" (_a6)
+#define LOAD_ARGS_1                                     \
+  register int _a1 __asm__ ("D1Ar1") = (int) (_t1);     \
+  LOAD_ARGS_0
+#define ASM_ARGS_1 ASM_ARGS_0, "d" (_a1)
+#define LOAD_ARGS_2                                     \
+  register int _a2 __asm__ ("D0Ar2") = (int) (_t2);     \
+  LOAD_ARGS_1
+#define ASM_ARGS_2 ASM_ARGS_1, "d" (_a2)
+#define LOAD_ARGS_3                                     \
+  register int _a3 __asm__ ("D1Ar3") = (int) (_t3);     \
+  LOAD_ARGS_2
+#define ASM_ARGS_3 ASM_ARGS_2, "d" (_a3)
+#define LOAD_ARGS_4                                     \
+  register int _a4 __asm__ ("D0Ar4") = (int) (_t4);     \
+  LOAD_ARGS_3
+#define ASM_ARGS_4 ASM_ARGS_3, "d" (_a4)
+#define LOAD_ARGS_5                                     \
+  register int _a5 __asm__ ("D1Ar5") = (int) (_t5);     \
+  LOAD_ARGS_4
+#define ASM_ARGS_5 ASM_ARGS_4, "d" (_a5)
+#define LOAD_ARGS_6                                     \
+  register int _a6 __asm__ ("D0Ar6") = (int) (_t6);     \
+  LOAD_ARGS_5
+#define ASM_ARGS_6 ASM_ARGS_5, "d" (_a6)
 
 #endif /* __ASSEMBLER__ */
 #endif /* _BITS_SYSCALLS_H */
diff --git a/libc/sysdeps/linux/metag/clone.S b/libc/sysdeps/linux/metag/clone.S
index 8fff56710..d9d836338 100644
--- a/libc/sysdeps/linux/metag/clone.S
+++ b/libc/sysdeps/linux/metag/clone.S
@@ -8,7 +8,17 @@
 #include <asm/errno.h>
 #include <asm/unistd.h>
 
-/* int clone(int (*fn)(void *arg), void *child_stack, int flags, void *arg); */
+#define CLONE_VM      0x00000100
+#define CLONE_THREAD  0x00010000
+
+#ifdef __PIC__
+#define __CLONE_METAG_LOAD_TP ___metag_load_tp@PLT
+#else
+#define __CLONE_METAG_LOAD_TP ___metag_load_tp
+#endif
+
+/* int clone(int (*fn)(void *arg), void *child_stack, int flags, void *arg,
+			pid_t *ptid, struct user_desc *tls, pid_t *ctid); */
 
 	.text
 	.global	__clone
@@ -25,8 +35,12 @@ __clone:
 	MOV D0FrT, D1Ar1
 
 	! do the system call
-	! get flags
 	MOV	D1Ar1, D1Ar3
+	MOV	D1Ar3, D1Ar5
+	MOV	D1Ar5, D0Ar6
+	MOV	D0Ar6, D0Ar4
+	GETD	D0Ar4, [A0StP+#-4]
+
 	! new sp is already in D0Ar2
 	MOV	D1Re0, #__NR_clone
 	SWITCH	#0x440001
@@ -38,14 +52,36 @@ __clone:
 	! BRKPNT
 
 	! We are the child
-	! pick the function arg and call address off the stack and execute
-	MOV     D0Ar2, D0FrT
-	MOV     D1Ar1, D0Ar4
-	MOV     D1RtP, PC
+#ifdef RESET_PID
+	SETL	[A0StP++], D0FrT, D1RtP
+	MOVT	D0FrT, #HI(CLONE_THREAD)
+	ADD	D0FrT, D0FrT, #LO(CLONE_THREAD)
+	ANDS	D0FrT, D0FrT, D1Ar1
+	BNZ	3f
+	MOVT	D0FrT, #HI(CLONE_VM)
+	ADD	D0FrT, D0FrT, #LO(CLONE_VM)
+	ANDS	D0FrT, D0FrT, D1Ar1
+	BZ	1f
+	MOV	D1Ar1, #-1
+	BA	2f
+1:	MOV	D1Re0, #__NR_getpid